A loss calculation method, device and equipment of a direct current power transmission system and a storage medium

Abstract

The application relates to a direct current power transmission system loss calculation method, device, equipment and storage medium, the method comprising: based on a full construction mask mechanism and an encoder-decoder architecture, an initial bidirectional mapping digital twin loss model is established; according to a migration algorithm, the initial bidirectional mapping digital twin loss model is trained and optimized to obtain a transition bidirectional mapping digital twin loss model; based on a knowledge distillation algorithm, the transition bidirectional mapping digital twin loss model is lightened to obtain a target bidirectional mapping digital twin loss model; and according to the target bidirectional mapping digital twin loss model, the loss of an actual direct current power transmission system is calculated. The application provides a direct current power transmission system loss calculation method, device, equipment and storage medium, bidirectional mapping is realized through a full construction mask mechanism and an encoder-decoder architecture, and the calculation accuracy and feature extraction capability are improved through a migration algorithm and a knowledge distillation algorithm.

Description

Technical Field

[0001] This invention relates to the field of power transmission system loss calculation technology, and in particular to a method, apparatus, equipment and storage medium for calculating losses in a DC power transmission system. Background Technology

[0002] Ultra-high voltage direct current (UHVDC) transmission technology can achieve regional power absorption, reduce short-circuit current levels, and effectively solve the power flow reversal problem in AC transmission technology, making it a primary mode of long-distance power transmission. However, UHVDC transmission systems typically incur converter station line losses and DC transmission line corona losses, leading to discrepancies in power settlement at the point of entry, affecting the fairness of power system transactions, and increasing the operating costs of power grid companies. Therefore, to provide a basis for energy conservation and loss reduction in UHVDC transmission systems and improve energy transmission efficiency, it is necessary to conduct research on loss measurement in UHVDC transmission systems.

[0003] Ultra-high voltage direct current (UHVDC) transmission systems are complex, multi-input, multi-output, nonlinear, and strongly coupled transmission systems. In existing technologies, to obtain more accurate calculation results, refined modeling and digital twins are commonly used for component-level loss studies. In refined modeling, transmission system loss measurement is mainly based on equivalent models established using the physical characteristics of loss-producing components and grid operating parameters. This modeling allows for in-depth loss analysis within the components, yielding high-precision simulation results. Digital twins offer unprecedented advantages, including timely detection capabilities, robustness under extreme conditions, and optimized maintenance strategies.

[0004] However, refined modeling has the problem of excessively high time and computational costs; the digital twin approach still has three drawbacks: 1) Traditional deep learning models have insufficient feature extraction capabilities and cannot complete the topological feature extraction of a large number of devices, thus failing to meet the inference accuracy requirements of digital twins; 2) Traditional deep learning models and Transformer models can only establish one-way mappings, which cannot meet the two-way mapping requirements of digital twins; 3) Although the Transformer model has made significant improvements in speed and accuracy, it still cannot meet the real-time requirements in digital twins. Summary of the Invention

[0005] In view of this, it is necessary to provide a method, apparatus, equipment and storage medium for calculating losses in DC transmission systems, in order to solve the problems of insufficient feature extraction capability of deep learning models in the prior art, inability to achieve bidirectional mapping and inaccurate loss calculation.

[0006] To achieve the above-mentioned technical objectives, the present invention adopts the following technical solution:

[0007] In a first aspect, the present invention provides a method for calculating losses in a DC transmission system, comprising:

[0008] Sample data from a DC transmission system were acquired, and an initial bidirectional mapping digital twin loss model was established based on a full-construction masking mechanism and an encoder-decoder architecture.

[0009] The transitional bidirectional mapping digital twin loss model is obtained by training and optimizing the initial bidirectional mapping digital twin loss model using the transfer algorithm.

[0010] Based on the knowledge distillation algorithm, a lightweight modification of the transition bidirectional mapping digital twin loss model is performed to obtain the target bidirectional mapping digital twin loss model.

[0011] The losses of a real DC transmission system are calculated based on a target bidirectional mapping digital twin loss model.

[0012] Preferably, sample data from the DC transmission system is acquired, and an initial bidirectional mapped digital twin loss model is established based on a full construction masking mechanism and an encoder-decoder architecture, including:

[0013] Establish a digital twin model based on an encoder-decoder architecture;

[0014] The sample data is masked using a full construction masking mechanism to obtain a mask sequence.

[0015] The initial bidirectional mapped digital twin loss model is obtained by training the digital twin model using a mask sequence through bidirectional mapping.

[0016] Preferably, a mask sequence is obtained by masking the sample data based on a full construction masking mechanism, including:

[0017] Valid data are randomly selected from the sample data of the DC transmission system, and a mask is assigned to the valid data to obtain a mask sequence.

[0018] Preferably, the digital twin model is trained by bidirectional mapping based on the mask sequence to obtain an initial bidirectional mapped digital twin loss model, including:

[0019] Based on the encoder-decoder architecture, the corresponding mapping data is determined according to the mask sequence;

[0020] Based on a preset loss function, the bidirectional mapping loss value is determined according to the mask data and the mapping data.

[0021] Based on the backpropagation algorithm, the digital twin model is bidirectionally mapped between the digital space and the physical space according to the bidirectional mapping loss value to obtain the initial bidirectional mapping digital twin loss model.

[0022] Preferably, a transitional bidirectional mapped digital twin loss model is obtained by training and optimizing the initial bidirectional mapped digital twin loss model using a transfer algorithm, including:

[0023] Acquire the DC transmission system dataset and set the number of iterations;

[0024] Based on the transfer learning algorithm, the initial bidirectional mapping digital twin loss model is trained and optimized according to the DC transmission system dataset and the number of cycles to obtain the transition bidirectional mapping digital twin loss model.

[0025] Preferably, based on the knowledge distillation algorithm, a lightweight modification is performed on the transitional bidirectional mapping digital twin loss model to obtain the target bidirectional mapping digital twin loss model, including:

[0026] A teacher model and a student model are established based on the knowledge distillation algorithm. The hidden layer feature vector of the teacher model is used to guide the student model to extract hidden layer features, thereby obtaining the feature vectors of the teacher model and the student model.

[0027] Through knowledge transfer, the loss value of the student model is determined based on the feature vectors of the teacher model and the student model.

[0028] The target bidirectional mapping digital twin loss model is obtained by lightweighting the transition bidirectional mapping digital twin loss model based on the loss value of the student model.

[0029] Preferably, through knowledge transfer, the loss value of the student model is determined based on the feature vectors of the teacher model and the student model, including:

[0030] The influence factors of the encoder loss and the hidden layer loss on the student model are determined based on the feature vectors of the teacher model and the features of the student model.

[0031] The loss value of the student model is determined based on the influence factors of the encoder loss on the student model and the influence factors of the hidden layer loss on the student model.

[0032] Secondly, the present invention also provides a DC transmission system loss calculation device, comprising:

[0033] The modeling module is used to acquire sample data of the DC transmission system and establish an initial bidirectional mapping digital twin loss model based on the full construction mask mechanism and encoder-decoder architecture.

[0034] The training module is used to train and optimize the initial bidirectional mapping digital twin loss model according to the transfer algorithm to obtain the transitional bidirectional mapping digital twin loss model.

[0035] The modification module is used to perform a lightweight modification on the transitional bidirectional mapping digital twin loss model based on the knowledge distillation algorithm to obtain the target bidirectional mapping digital twin loss model.

[0036] The calculation module is used to calculate the losses of the actual DC transmission system based on the target bidirectional mapping digital twin loss model.

[0037] Thirdly, the present invention also provides an electronic device, including a memory and a processor, wherein,

[0038] Memory, used to store programs;

[0039] The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the DC transmission system loss calculation method in any of the above implementations.

[0040] Fourthly, the present invention also provides a computer-readable storage medium for storing a computer-readable program or instruction, which, when executed by a processor, can implement the steps in the DC transmission system loss calculation method described in any of the above implementations.

[0041] The beneficial effects of the above embodiments are as follows: The DC transmission system loss calculation method, apparatus, equipment, and storage medium provided by the present invention establish an initial bidirectional mapping digital twin loss model based on a full construction mask mechanism and encoder-decoder architecture, realizing bidirectional mapping between digital space and physical space, and meeting the bidirectional mapping requirements of digital twins. The initial bidirectional mapping digital twin loss model is trained and optimized using a transfer algorithm to obtain a transitional bidirectional mapping digital twin loss model. Applying the digital space bidirectional mapping digital twin model to loss measurement enables accurate application of the bidirectional mapping digital twin loss model in loss measurement. Furthermore, based on a knowledge distillation algorithm, the transitional bidirectional mapping digital twin loss model is lightweighted to obtain the target bidirectional mapping digital twin loss model. Knowledge distillation achieves the effect of refining the model, significantly improving model efficiency within the allowable range of deduction accuracy, and enhancing the model's feature extraction capability. Attached Figure Description

[0042] Figure 1 A flowchart illustrating an embodiment of the DC transmission system loss calculation method provided by the present invention;

[0043] Figure 2 for Figure 1 A schematic flowchart of an embodiment of step S101;

[0044] Figure 3 for Figure 2 A flowchart illustrating an embodiment of step S203;

[0045] Figure 4 for Figure 1 A schematic flowchart of an embodiment of step S103;

[0046] Figure 5 A schematic diagram of an embodiment of the DC transmission system loss calculation device provided by the present invention;

[0047] Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0048] Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, which form part of this application and are used together with the embodiments of the present invention to illustrate the principles of the present invention, but are not intended to limit the scope of the present invention.

[0049] In the description of this application, "multiple" means two or more, unless otherwise expressly and specifically defined.

[0050] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0051] This invention provides a method, apparatus, equipment, and storage medium for calculating losses in a DC transmission system, which will be described below.

[0052] Please see Figure 1 , Figure 1 This is a flowchart illustrating an embodiment of the DC transmission system loss calculation method provided by the present invention. A specific embodiment of the present invention discloses a DC transmission system loss calculation method, comprising:

[0053] S101. Obtain sample data of the DC transmission system and establish an initial bidirectional mapping digital twin loss model based on the full construction mask mechanism and encoder-decoder architecture.

[0054] S102. The initial bidirectional mapping digital twin loss model is trained and optimized according to the transfer algorithm to obtain the transitional bidirectional mapping digital twin loss model.

[0055] S103. Based on the knowledge distillation algorithm, the transition bidirectional mapping digital twin loss model is lightened and modified to obtain the target bidirectional mapping digital twin loss model.

[0056] S104. Calculate the losses of the actual DC transmission system based on the target bidirectional mapping digital twin loss model.

[0057] In the above embodiments, the DC transmission system in this application is an ultra-high voltage DC transmission system, where ultra-high voltage refers to DC power of ±800 kV and above.

[0058] The bidirectional mapping consists of two parts: forward deduction and reverse inversion. The forward deduction part uses an encoding and decoding structure to complete the mapping from the digital space to the physical space of the UHVDC transmission system; the reverse inversion part uses a full-construction masking mechanism to complete the reverse mapping from the physical space to the digital space.

[0059] The initial bidirectional mapped digital twin loss model cannot calculate the losses of UHVDC transmission systems. By applying the initial bidirectional mapped digital twin loss model to loss metering through the transfer algorithm, the bidirectional mapped digital twin loss model can be accurately applied to the loss metering function. The resulting transitional bidirectional mapped digital twin loss model can then calculate the losses of UHVDC transmission systems, but its calculation accuracy still needs to be further improved.

[0060] Knowledge distillation algorithms can improve the feature extraction capability of bidirectional mapping digital twin loss models. By modifying the transitional bidirectional mapping digital twin loss model using knowledge distillation algorithms, the feature extraction capability of the bidirectional mapping digital twin loss model is improved to obtain the target bidirectional mapping digital twin loss model, which in turn improves the accuracy of loss calculation for UHVDC transmission systems.

[0061] Compared with existing technologies, this embodiment provides a DC transmission system loss calculation method. Based on a fully constructed masking mechanism and an encoder-decoder architecture, an initial bidirectional mapping digital twin loss model is established, realizing bidirectional mapping between digital space and physical space, thus meeting the bidirectional mapping requirements of digital twins. The initial bidirectional mapping digital twin loss model is trained and optimized using a transfer algorithm to obtain a transitional bidirectional mapping digital twin loss model. Applying the bidirectional mapping digital twin model to loss measurement enables accurate application of the bidirectional mapping digital twin loss model in loss measurement. Then, based on a knowledge distillation algorithm, the transitional bidirectional mapping digital twin loss model is lightweighted to obtain the target bidirectional mapping digital twin loss model. Knowledge distillation achieves the effect of refining the model, significantly improving model efficiency and enhancing feature extraction capabilities within the allowable range of deduction accuracy.

[0062] Please see Figure 2 , Figure 2 for Figure 1 A flowchart illustrating an embodiment of step S101. In some embodiments of the present invention, sample data of a DC transmission system is acquired, and an initial bidirectional mapped digital twin loss model is established based on a full construction mask mechanism and an encoder-decoder architecture, including:

[0063] S201. Establish a digital twin model based on the encoder-decoder architecture;

[0064] S202. Masking the sample data based on the full construction masking mechanism to obtain the mask sequence;

[0065] S203. Based on the mask sequence, perform bidirectional mapping training on the digital twin model to obtain the initial bidirectional mapping digital twin loss model.

[0066] In the above embodiments, the initial data, i.e., sample data of the DC transmission system, is obtained by acquiring the operational data sequence of the actual UHVDC transmission system, and then a digital twin model is directly established. It should be noted that the digital twin model is existing technology, and the process of establishing the digital twin model will not be elaborated upon here.

[0067] The sample data of the DC transmission system is processed by a full-construction masking mechanism and an encoder-decoder architecture, and the digital twin model is trained by bidirectional mapping. Based on the bidirectional mapping model parameters and the digital twin model, the digital twin model is processed by forward deduction and inverse inversion to obtain the initial bidirectional mapping digital twin loss model, so as to realize the bidirectional mapping between digital space and physical space.

[0068] In some embodiments of the present invention, a mask sequence is obtained by masking sample data based on a full construction masking mechanism, including:

[0069] Valid data are randomly selected from the sample data of the DC transmission system, and a mask is assigned to the valid data to obtain a mask sequence.

[0070] In the above embodiments, a full construction masking mechanism is applied to the sequence (sample data of the DC transmission system). The principle of randomly selecting valid data from the data sequence is used to assign a mask to the input digital space prior bidirectional mapping data to obtain masked data. In other words, the data x at position i is... i Replace it with a mask to keep the length of the entire data sequence unchanged, only obscuring part of the data.

[0071] Please see Figure 3 , Figure 3 for Figure 2 A flowchart illustrating an embodiment of step S203. In some embodiments of the present invention, a bidirectional mapping training is performed on the digital twin model based on a mask sequence to obtain an initial bidirectional mapping digital twin loss model, including:

[0072] S301. Based on the encoder-decoder architecture, determine the corresponding mapping data according to the mask sequence;

[0073] S302. Based on the preset loss function, determine the bidirectional mapping loss value according to the mask data and mapping data;

[0074] S303. Based on the backpropagation algorithm, the digital twin model is bidirectionally mapped between the digital space and the physical space according to the bidirectional mapping loss value to obtain the initial bidirectional mapping digital twin loss model.

[0075] In the above embodiment, the mask sequence is used as the input of the encoder-decoder structure. The encoder-decoder structure calculates the value of the masked data, completes the mapping calculation from the mask data to the true value xi data, and obtains the model's mapping output x′i, i.e., the mapping data.

[0076] The default loss function is:

[0077]

[0078] Where m is the number of mask data. This represents the bidirectional mapping loss value.

[0079] The mask data and mapping data are used as inputs, and the bidirectional mapping loss value is calculated according to the preset loss function.

[0080] Based on the loss value, the backpropagation algorithm is used to process the digital twin model by bidirectional mapping of the loss value, ultimately realizing the bidirectional mapping between the digital space and the physical space of the UHVDC transmission system.

[0081] In some embodiments of the present invention, a transitional bidirectional mapped digital twin loss model is obtained by training and optimizing the initial bidirectional mapped digital twin loss model according to a transfer algorithm, including:

[0082] Acquire the DC transmission system dataset and set the number of iterations;

[0083] Based on the transfer learning algorithm, the initial bidirectional mapping digital twin loss model is trained and optimized according to the DC transmission system dataset and the number of cycles to obtain the transition bidirectional mapping digital twin loss model.

[0084] In the above embodiments, a DC transmission system dataset is obtained based on actual UHVDC transmission system data sequences. For loss transfer applications, dedicated training and testing datasets need to be designed. Within the DC transmission system dataset, n data sets are randomly selected from each set. mask The mask data is calculated using the following formula:

[0085] n mask =L×r mask

[0086] Where the length of each data set is L, r mask n is the mask ratio. mask The number of mask data is used. The dataset is expanded by randomly adding masks multiple times.

[0087] The number of iterations is set, and the initial bidirectional mapping digital twin loss model is trained and optimized multiple times based on the DC transmission system dataset until the set number of iterations is reached. After multiple transfer trainings, the transition bidirectional mapping digital twin loss model is accurately applied to the metering loss function.

[0088] Please see Figure 4 , Figure 4 for Figure 1 A flowchart illustrating one embodiment of step S103. In some embodiments of the present invention, based on a knowledge distillation algorithm, a lightweight modification is performed on the transitional bidirectional mapping digital twin loss model to obtain the target bidirectional mapping digital twin loss model, including:

[0089] S401. Based on the knowledge distillation algorithm, establish teacher and student models, and use the hidden layer feature vector of the teacher model to guide the student model to extract hidden layer features to obtain the feature vectors of the teacher model and the student model.

[0090] S402. Through knowledge transfer, determine the loss value of the student model based on the feature vectors of the teacher model and the student model.

[0091] S403. Based on the loss value of the student model, the transitional bidirectional mapping digital twin loss model is lightened to obtain the target bidirectional mapping digital twin loss model.

[0092] In the above embodiment, the hidden layer feature vector of the teacher model is known. Based on the hidden layer feature vector of the teacher model, the student model is guided to extract hidden layer features, realize feature learning, and obtain the feature vector of the student model.

[0093] Then, the teacher model guides the student model to learn the device topology coding matrix, enabling knowledge transfer from the teacher model to the student model. During this knowledge transfer process, the student model's loss function L... student The calculation formula is:

[0094] L student =L pred +λ1L encoder +λ2L hidden ;

[0095] Among them, L pred L is the deviation between the results obtained by the student model using digital space data and the true values. encoder It is the encoder loss between the student model and the teacher model, L hidden λ1 and λ2 are the hidden layer loss between the student model and the teacher model, and the weight control coefficients are λ1 and λ2.

[0096] In some embodiments of the present invention, the loss value of the student model is determined based on the feature vectors of the teacher model and the student model through knowledge transfer, including:

[0097] The influence factors of the encoder loss and the hidden layer loss on the student model are determined based on the feature vectors of the teacher model and the features of the student model.

[0098] The loss value of the student model is determined based on the influence factors of the encoder loss on the student model and the influence factors of the hidden layer loss on the student model.

[0099] In the above embodiments, the impact of encoder loss calculation on the student model can be described as follows:

[0100] L encoder =MSE(F S F T );

[0101] Among them, F T and F S These represent the output feature vectors of the encoder in the teacher model and the student model, respectively.

[0102] The impact of calculating the loss for the hidden layer on the student model can be described as follows:

[0103]

[0104] The nth layer sequence features represent the teacher model. The m-th layer sequence features represent the student model.

[0105] This invention also uses the control variable method to measure the deviation between the actual and calculated values ​​of losses in ultra-high voltage direct current transmission systems, using indicators such as mean square error (MSE), root mean square error (RMSE), mean absolute error (MAE), logarithmic error of mean square error (MSLE), and R² score. An ablation experiment was conducted on the proposed target bidirectional mapping digital twin loss model.

[0106] First, derivation experiments were conducted on the three independent improvement strategies. Notably, on the bidirectional mapping dataset, topological feature extraction, device location encoding, and full construction masking mechanisms were all used in the forward derivation. Subsequently, the three improvement strategies were gradually combined and derivation experiments were conducted sequentially, ultimately resulting in eight sets of ablation experiments, as shown in Table 1.

[0107] Table 1

[0108]

[0109]

[0110] The experimental results in the table show that topology feature extraction, device location encoding, and full construction mask all have poor accuracy in independent forward inference. Comparing Experiments 1, 2, 3, and 4, it can be seen that when using the strategy alone, device location encoding contributes the most to the inference, followed by the full construction mask, while topology feature extraction provides the least improvement. However, the MSE error is still reduced by 54.27% compared to the case without the mask, and the R2 score is also improved. Furthermore, observing the contribution of the full construction mask to the model, comparing Experiments 3 and 7, the full construction mask reduces the MSE loss by 53.56%, and the MSE loss in Experiments 2 and 6 is reduced by 58.31%. This shows that the full construction mask can further enhance the model's inference ability. However, it is undeniable that the addition of the full construction mask will sacrifice some stability; for example, both sets of comparisons show a loss of 0.30% and 0.76% in the R2 score. Finally, Experiment 8, using the complete improvement strategy, achieved a minimum reduction of 16.71% in MSE loss compared to other ablation groups (Experiment 4). However, the R2 score was not the highest. This is because the addition of the full construction mask caused the model to lose some data information during learning, which is also the reason for the improved accuracy.

[0111] Furthermore, three mainstream neural networks were selected and optimized to achieve the optimal state for digital twin loss measurement extrapolation of UHVDC transmission systems. Then, loss extrapolation was performed using a transfer application dataset, and the digital twin model (DTformer) was compared with other similar algorithms. The results are shown in Table 2.

[0112] Table 2

[0113]

[0114] As can be seen, CNN networks, which excel at feature extraction in the image domain, cannot effectively perform loss estimation for UHVDC transmission systems. Among traditional deep learning models, RNNs, due to their ability to extract temporal features, reduce the MSE error by 49.14% compared to CNNs using convolutional features, indicating that temporal feature extraction models are more suitable for UHVDC digital twin estimation. BiLSTM, using inter-layer information transfer and a bidirectional strategy, reduces the MSE error by 73.00% compared to RNNs. Finally, the Transformer model, based on equipment location encoding and topological feature extraction, further reduces the estimation error. Based on this, we propose the DTformer model, which is more suitable for UHVDC transmission systems. It can be seen that the DTformer model is superior to mainstream neural network models in loss estimation.

[0115] Furthermore, a smaller DTformer-mini model was constructed. The number of encoders and decoders was reduced to two, the number of channels and dimensions of the main device topology feature extraction module was reduced, and the number of network layers was reduced by 50%. Module lightweighting experiments were conducted on the DTformer digital twin model, including pyramid sampling lightweighting and integrated feature channel lightweighting. A comparative experiment on the lightweighting of loss metering migration applications of the DTformer digital twin model was also carried out, as shown in Table 3.

[0116] Table 3

[0117]

[0118] Analysis of the experimental results in Table 3 shows that the unprocessed DTformer model has the best inference accuracy, but also the longest inference time. In the four comparative experiments without distillation techniques, the MSE error of the model continuously increases with the use of various lightweighting techniques, while the inference time decreases, which is consistent with the principle of lightweighting strategies. It can be seen that the acceleration capability of integrated feature channel lightweighting is stronger than that of pyramid sampling lightweighting, shortening the inference time by 0.019s, but the accuracy decreases by 16.79%. The student model using the mini structure is stronger than the previous two in terms of acceleration capability, with an inference speed increase of 7.97 times, but it sacrifices the most accuracy.

[0119] Observing the performance of distillation techniques, compared with not using distillation, DTformer's MSE error increased by 1.67 times, while the inference time was shortened by 0.30 seconds. Pyramid sampling lightweighting and integrated feature channel lightweighting also showed an increase in MSE error but an increase in inference speed. However, DTforner-mini, which uses a mini model and incorporates distillation, reduced the MSE error by 57.86% and increased the inference speed by 1.28 times. This result is because the distillation structure designed according to the student-teacher model perfectly fits DTformer-mini. The refined design reduces the model size while improving the information transmission efficiency of the encoder and decoder, allowing the mini model to improve both inference accuracy and speed. Ultimately, this millimeter-level response demonstrates excellent performance in digital twin loss inference for UHVDC.

[0120] This invention addresses the shortcomings of conventional deep learning models and modeling methods by proposing a novel modeling and application architecture for a target-oriented bidirectional mapping digital twin loss model in physical space. This architecture improves the accuracy, generalization ability, and inversion capability of the digital twin model. Specifically, it addresses the deficiencies of conventional deep learning models and modeling methods by enhancing the accuracy, ability to handle missing data, and inversion capability of the digital twin model. Experimental results show that all error metrics of the target-oriented bidirectional mapping digital twin loss model are significantly superior to conventional deep learning algorithms. Taking MSE error as an example, the maximum reduction is 92.55%, and the minimum is 63.79%. This significantly improves the speed, accuracy, and real-time performance of loss metering in ultra-high voltage direct current transmission systems.

[0121] To better implement the DC transmission system loss calculation method in this embodiment of the invention, based on the DC transmission system loss calculation method, please refer to the corresponding documentation. Figure 5 , Figure 5 This is a schematic diagram of an embodiment of the DC transmission system loss calculation device provided by the present invention. The embodiment of the present invention provides a DC transmission system loss calculation device 500, comprising:

[0122] Modeling module 510 is used to acquire sample data of DC transmission system and establish an initial bidirectional mapping digital twin loss model based on full construction masking mechanism and encoder-decoder architecture;

[0123] Training module 520 is used to train and optimize the initial bidirectional mapping digital twin loss model according to the transfer algorithm to obtain the transition bidirectional mapping digital twin loss model.

[0124] Modification module 530 is used to perform lightweight modification of the transitional bidirectional mapping digital twin loss model based on the knowledge distillation algorithm to obtain the target bidirectional mapping digital twin loss model.

[0125] The calculation module 540 is used to calculate the losses of the actual DC transmission system based on the target bidirectional mapping digital twin loss model.

[0126] It should be noted that the device 500 provided in the above embodiments can implement the technical solutions described in the above method embodiments. The specific implementation principles of the above modules or units can be found in the corresponding content in the above method embodiments, and will not be repeated here.

[0127] Please see Figure 6 , Figure 6This is a schematic diagram of the structure of an electronic device provided in an embodiment of the present invention. Based on the above-described method for calculating losses in a DC transmission system, the present invention also provides a DC transmission system loss calculation device, which can be a mobile terminal, desktop computer, laptop, handheld computer, server, or other computing device. The DC transmission system loss calculation device includes a processor 610, a memory 620, and a display 630. Figure 6 Only some components of the electronic device are shown; however, it should be understood that it is not required to implement all of the components shown, and more or fewer components may be implemented instead.

[0128] In some embodiments, memory 620 may be an internal storage unit of a DC transmission system loss calculation device, such as a hard disk or RAM of the DC transmission system loss calculation device. In other embodiments, memory 620 may be an external storage device of the DC transmission system loss calculation device, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the DC transmission system loss calculation device. Furthermore, memory 620 may include both internal and external storage units of the DC transmission system loss calculation device. Memory 620 is used to store application software and various types of data installed on the DC transmission system loss calculation device, such as the program code for installing the DC transmission system loss calculation device. Memory 620 may also be used to temporarily store data that has been output or will be output. In one embodiment, memory 620 stores a DC transmission system loss calculation program 640, which can be executed by processor 610 to implement the DC transmission system loss calculation method of the various embodiments of this application.

[0129] In some embodiments, processor 610 may be a central processing unit (CPU), microprocessor, or other data processing chip, used to run program code stored in memory 620 or process data, such as executing a method for calculating losses in a DC transmission system.

[0130] In some embodiments, display 630 may be an LED display, a liquid crystal display, a touch-screen liquid crystal display, or an OLED (Organic Light-Emitting Diode) touchscreen. Display 630 is used to display information from the DC transmission system loss calculation device and to display a user interface for visualization. Components 610-630 of the DC transmission system loss calculation device communicate with each other via a system bus.

[0131] In one embodiment, when the processor 610 executes the DC transmission system loss calculation program 640 in the memory 620, it implements the steps in the DC transmission system loss calculation method described above.

[0132] This embodiment also provides a computer-readable storage medium storing a DC transmission system loss calculation program thereon, which, when executed by a processor, performs the following steps:

[0133] Sample data from a DC transmission system were acquired, and an initial bidirectional mapping digital twin loss model was established based on a full-construction masking mechanism and an encoder-decoder architecture.

[0134] The transitional bidirectional mapping digital twin loss model is obtained by training and optimizing the initial bidirectional mapping digital twin loss model using the transfer algorithm.

[0135] Based on the knowledge distillation algorithm, a lightweight modification of the transition bidirectional mapping digital twin loss model is performed to obtain the target bidirectional mapping digital twin loss model.

[0136] The losses of a real DC transmission system are calculated based on a target bidirectional mapping digital twin loss model.

[0137] In summary, the DC transmission system loss calculation method, apparatus, equipment, and storage medium provided in this embodiment establish an initial bidirectional mapping digital twin loss model based on a fully constructed masking mechanism and an encoder-decoder architecture. This achieves bidirectional mapping between the digital space and the physical space, meeting the bidirectional mapping requirements of the digital twin. A transitional bidirectional mapping digital twin loss model is obtained by training and optimizing the initial bidirectional mapping digital twin loss model using a transfer algorithm. Applying the bidirectional mapping digital twin model to loss measurement enables accurate application of the bidirectional mapping digital twin loss model in loss measurement. Furthermore, based on a knowledge distillation algorithm, the transitional bidirectional mapping digital twin loss model is lightweighted to obtain the target bidirectional mapping digital twin loss model. Knowledge distillation achieves the effect of refining the model, significantly improving model efficiency and enhancing feature extraction capabilities within the allowable range of deduction accuracy.

[0138] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for calculating losses in a DC transmission system, characterized in that, include: Sample data from a DC transmission system were acquired, and an initial bidirectional mapping digital twin loss model was established based on a full-construction masking mechanism and an encoder-decoder architecture. The initial bidirectional mapping digital twin loss model is trained and optimized using a transfer algorithm to obtain a transitional bidirectional mapping digital twin loss model. Based on the knowledge distillation algorithm, the transitional bidirectional mapping digital twin loss model is lightened and modified to obtain the target bidirectional mapping digital twin loss model. The losses of the actual DC transmission system are calculated based on the target bidirectional mapping digital twin loss model. The acquisition of sample data from the DC transmission system, based on a full-construction masking mechanism and an encoder-decoder architecture, establishes an initial bidirectional mapped digital twin loss model, including: Establish a digital twin model based on an encoder-decoder architecture; The sample data is masked using a full construction masking mechanism to obtain a mask sequence. The digital twin model is trained by bidirectional mapping based on the mask sequence to obtain the initial bidirectional mapped digital twin loss model.

2. The method for calculating losses in a DC transmission system according to claim 1, characterized in that, The masking process for the sample data based on the full construction masking mechanism to obtain the mask sequence includes: Valid data is randomly selected from the sample data of the DC transmission system, and a mask is assigned to the valid data to obtain a mask sequence.

3. The method for calculating losses in a DC transmission system according to claim 2, characterized in that, Based on the mask sequence, a bidirectional mapping training is performed on the digital twin model to obtain the initial bidirectional mapping digital twin loss model, including: Based on the encoder-decoder architecture, the corresponding mapping data is determined according to the mask sequence; Based on a preset loss function, a bidirectional mapping loss value is determined according to the mask sequence and the mapping data; Based on the backpropagation algorithm, the digital twin model is bidirectionally mapped between the digital space and the physical space according to the bidirectional mapping loss value to obtain the initial bidirectional mapping digital twin loss model.

4. The method for calculating losses in a DC transmission system according to claim 1, characterized in that, The step of training and optimizing the initial bidirectional mapped digital twin loss model according to the transfer algorithm to obtain the transitional bidirectional mapped digital twin loss model includes: Acquire the DC transmission system dataset and set the number of iterations; Based on the aforementioned transfer algorithm, the initial bidirectional mapping digital twin loss model is trained and optimized according to the DC transmission system dataset and the number of cycles to obtain the transitional bidirectional mapping digital twin loss model.

5. The method for calculating losses in a DC transmission system according to claim 1, characterized in that, The step of using a knowledge distillation algorithm to perform a lightweight modification on the transitional bidirectional mapping digital twin loss model to obtain the target bidirectional mapping digital twin loss model includes: Based on the knowledge distillation algorithm, a teacher model and a student model are established. The hidden layer feature vector of the teacher model is used to guide the student model to extract hidden layer features, thereby obtaining the feature vectors of the teacher model and the student model. Through knowledge transfer, the loss value of the student model is determined based on the feature vectors of the teacher model and the student model. The target bidirectional mapping digital twin loss model is obtained by lightweighting the transition bidirectional mapping digital twin loss model based on the loss value of the student model.

6. The method for calculating losses in a DC transmission system according to claim 5, characterized in that, The step of determining the loss value of the student model through knowledge transfer based on the feature vectors of the teacher model and the student model includes: The influence factors of the encoder loss on the student model and the influence factors of the hidden layer loss on the student model are determined based on the feature vector of the teacher model and the features of the student model. The loss value of the student model is determined based on the influence factors of the encoder loss on the student model and the influence factors of the hidden layer loss on the student model.

7. A loss calculation device for a DC transmission system, characterized in that, A method for calculating the loss of a DC transmission system according to any one of claims 1-6 includes: The modeling module is used to acquire sample data of the DC transmission system and establish an initial bidirectional mapping digital twin loss model based on the full construction mask mechanism and encoder-decoder architecture. The training module is used to train and optimize the initial bidirectional mapping digital twin loss model according to the transfer algorithm to obtain the transitional bidirectional mapping digital twin loss model. The modification module is used to perform a lightweight modification on the transitional bidirectional mapping digital twin loss model based on the knowledge distillation algorithm to obtain the target bidirectional mapping digital twin loss model. The calculation module is used to calculate the losses of the actual DC transmission system based on the target bidirectional mapping digital twin loss model.

8. An electronic device, characterized in that, Including memory and processor, among which, The memory is used to store programs; The processor, coupled to the memory, is used to execute the program stored in the memory to implement the steps in the DC transmission system loss calculation method according to any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, Used to store computer-readable programs or instructions, which, when executed by a processor, are capable of implementing the steps in the DC transmission system loss calculation method according to any one of claims 1 to 6.

Images